← back to timeline

EXP-DOSE100

artifact

Integrative Dose-Response — 100% (5/5 Synthesizers, Pure Position-Driven)

2026-03-10 L2 level paper3 4 runs $0.02
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

All five agents receive identical instructions — differing only in names. If hierarchy still emerges, it is generated by the turn-order structure itself (who speaks first/last), not by differences in personalities. This is a fundamental test: is hierarchy a property of structure or of content?

What we found

No predictions scored

Predictions we made before running

0/4 confirmed

Technical details

Research hypothesis

DOSE-RESPONSE SERIES: 0% → 20% → 40% → 60% → 80% → [100%]

100% = ALL agents are synthesizers. NO persona diversity whatsoever. All 5 agents receive identical instructions (except name).

This is the PURE POSITION TEST: any hierarchy that emerges can ONLY come from speaking order position (round-robin scheduling). There are zero persona differences to drive role differentiation.

Key question: What is the "floor" of position-driven hierarchy?

Three competing predictions: (a) Damped oscillation predicts: small negative swing from 80%, TR ≈ 0.33-0.36. The attractor is around 0.35. (b) FULL COLLAPSE: identical agents → identical behavior → flat hierarchy → TR ≈ 0. No persona variation means no reason for different vocabulary adoption rates. (c) POSITION CREATES HIERARCHY: even with identical agents, first-mover advantage and last-mover absorption create a non-zero gradient. TR ≈ 0.20-0.30. This would prove that SPEAKING ORDER ALONE generates hierarchy in round-robin.

This is the most theoretically important point in the dose-response curve. If TR > 0 at 100%, hierarchy is STRUCTURAL (position-driven), not persona-driven. If TR ≈ 0, hierarchy requires persona diversity.

Varied rotations [0,1,2,3] test whether individual TR assignments simply rotate with speaking order (pure position artifact) or have additional stochastic variation.

Experimental setup

Type: single_condition

Condition Parameters
PERSONA_LIVE persona: PERSONA, interaction: LIVE

Factors: persona (PERSONA) × interaction (LIVE)

Parameters

n_agents
5
n_rounds
50
n_runs_per_condition
4
model
gemini-2.5-flash-lite
temperature
0.9
scheduler
round_robin

Trophic Ratios by Condition

Mean trophic ratio per agent across runs. Error bars = ±1 std dev. Higher TR = more upstream (exporter).